Artificial Intelligence Billions Invested in Technology with an Expiry Date

Related Vendors

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

Technology companies are investing huge sums of money in an AI infrastructure whose value will erode in three to five years because hardware cycles are extremely short. This has little to do with investing in sustainable technology.

The global AI euphoria has triggered a historic wave of investment from which large tech companies such as Nvidia are known to be benefiting. According to an analysis by the Wall Street Journal, technology companies have invested more money in data centers, chips and power supplies for artificial intelligence in just three years than the US has spent on building its interstate highway system in four decades (via Trendforce).

Thanks to technological advances and a growing provider landscape, artificial intelligence has long since moved on from the "it's just a bubble" comments of previous years. AI is here to stay. What could be debated, however, is whether the valuation of some companies such as Nvidia is itself a bubble. However, even a fall in the share price today would have less impact than a few years ago, as competitors such as AMD and hyperscalers such as Google and Amazon have caught up with their own AI chips. However, the huge sums of money being invested in new data centers could prove to be a risky bet, and not just in terms of the profitability of AI offerings.

Problems with Outdated Hardware

Analysts assume that most AI processors will be technically obsolete after three to five years, as the development cycles for chips and AI accelerators are now quite short. In 12 to 18 months, the technology will take a significant step towards greater computing power and higher energy efficiency. This means that the value of the data centers themselves is rapidly eroding.

This effect is being accelerated by ever new chip generations, increasing efficiency requirements, changing AI architectures and high energy requirements, which quickly make older systems economically unviable. At the same time, the thermal and electrical load on the hardware in continuous operation also limits the physical service life of a chip.

And the Profit?

According to calculations by the Wall Street Journal, consumers and companies would have to invest around USD 800 billion in AI products over the entire lifespan of the data centers currently being built in order to justify the massive investments.

The discrepancy is particularly clear with OpenAI: according to the report, the ChatGPT company is said to have committed to paying Oracle around USD 60 billion per year for server services, with an expected turnover of USD 13 billion in 2025. Analysts assume that the global AI infrastructure will require annual revenues of around USD 2 trillion by 2030 in order to become sustainably profitable. By comparison, this exceeds the combined annual revenues of Amazon, Apple, Alphabet, Microsoft, Meta and Nvidia in 2024.

Bloomberg's analysts note that AI companies are, after all, experiencing rapid sales growth: The expected USD 13 billion in sales anticipated for OpenAI represents a tripling of previous sales. Competitor Anthropic already generates more than 5 billion US dollars annually. However, according to the authors of Bloomberg, these figures pale in comparison to the expected cost increases. (sb)